A New Component Scattering Model Using Polarimetric Signatures Based Pattern Recognition on Polarimetric SAR Data

Abstract

This paper presents a new method for reconstruction of scattering mechanism contributions in quad-polarimetric SAR (PolSAR) data. Scattering mechanisms of one pixel are reconstructed using a comparison among polarimetric signatures of this pixel with polarimetric signatures of four canonical objects including trihedral (sphere or flat plate), dihedral, helix, and dipole. Pattern recognition matching methods are utilized in this comparison. In this research, a full polarimetric Radarsat-2 image was chosen for evaluation of the proposed method. The study area included various land cover classes e.g., different forest species, urban, water, and ground vegetation. The eight features corresponding to four canonical objects and two signatures (co-polarized and cross-polarized) provided by the proposed method were analyzed in various classes. The results of SVM classifier using these features were compared with results obtained from SVM classifiers using features provided by Freeman, Van Zyl, and Yamaguchi decomposition methods. Results showed that proposed feature set extracts new concepts from the images which are different from the concept extracting by other features presented in previous studies. Also, the accuracy of the proposed method in recognition of forest species was better than the other methods.